David Tortorian, a recent graduate of MIT and a member of the technical staff at the Startup Lab Somatic superior intelligence (PSI)use early access to Anthropic’s Fable 5.2 model to solve the problem Eighteen year old problem About quantum computing.
“I completely accept that this may be the last paper I ever release,” Tortorian, who postponed his plans to pursue a master’s degree at MIT’s prestigious Computer Science and Artificial Intelligence Laboratory to join PSI, told TechCrunch. “In a few months, why would you even issue a paper?”
The problem in question, the Aronson-Ambainis conjecture, concerns the ability of quantum computers to solve problems faster than classical ones. Scott Aaronson, one of the researchers who posed the problem, He writes He doesn’t fully understand the proof, but he believes it’s correct because it was automatically formulated in Lean, a mathematics markup language, and He explained By a team of senior researchers with whom PSI shared the solution.
The solution appears to be closing the door on one way quantum computers could solve some, but not all, difficult problems at high speed. “It’s a negative result that lets people know, ‘OK, we’re not looking there anymore,'” Tortorian said.
More broadly, it reflects the disruptions in the scientific world caused by problem solving using artificial intelligence.
“Proof of the complete [Aaronson-Ambainis conjecture] “It was one of the last remaining open problems in the complexity of quantitative inquiry that everyone seemed to agree was of central importance,” wrote Aronson, who described himself feeling satisfied, awed, disappointed, and exhausted by the news. “I was surprised, frankly, to see it missing from OpenAI’s Mathocalypse repository a few weeks ago. Sure, they tried, so was it that much more difficult than all the problems they solved?”
This is a point in favor of the PSI model, which aims to solve scientific problems using specialized models and scaffolds.
“An effective type of research system has to have a lot of instrument-related things, a lot of instrument-related things, memory systems, and simulation infrastructure,” PSI CEO Matt Baines told TechCrunch. “You want to be able to really go after a problem that is broad and deep, and be able to think of a wide range of potential alternatives.”
Currently, the company’s revenue comes from using its tools to optimize data centers, simulating heat, power, water and sound variables — “where the performance of that site as well as its impact on the community is a direct function of its physics.” However, the long-term task is to turn theoretical physics breakthroughs into actual technologies.
“Six to nine months ago, when I was talking to investors, like I had to kind of stretch to validate this condition, right? Like, no, no, no, trust me, very soon, theoretical physics is going to be mostly the domain of machines,” Baines said. “It’s not there yet, but it’s on its way.”
However, while AI has claimed important solutions in fields such as mathematics and computer science where they can be expressed in validated code, doing so in less abstract disciplines such as life sciences or physics has proven more difficult. Last week, startup Discovered Materials launched a $1 million mega-challenge for researchers using artificial intelligence to understand problems like superconductivity and heat dissipation.
“Ultimately, physics is different from mathematics because it’s the real world…nature has a vote in your models of reality,” Baines said. He said that the startup is focused on understanding how to obtain computational results and verify their validity through experiments in the most effective ways.
“If this is a real fact of nature, then this is a trick you can exploit,” Baines said. “You can build a device out of it, right? AI suggests an idea; it lets you do something in the world that you couldn’t do before…and so each one of these is a kind of loop that PSI is building.”
As the “New Laboratories of Science” become a pathway to potentially groundbreaking science, participants are still coming up with the parameters of this new world. Since PSI had to hold on to their results until the Anthropic model was almost ready for release, they were able to share them with the researchers involved to better understand it.
“As academics, we wanted to play our part in doing things the right way,” said Jim Halvorson, a professor at Northeastern University who advises PSI. He will attend a summit at MIT next week where leading physicists will try to come up with a response to the new pace of science. “Best practices are still emerging… We’ve found that transparency is a good way forward, and connecting with experts is a good way forward.”
“We had some time to sit on it… and it was the first time I had actually printed out an AI solution, read it, and thought, OK, how can I make this… present more naturally to a human?” Tortorian, who said the academic community expected him to be able to explain the result, told TechCrunch. “And I think that also made me realize, oh my gosh, the scale of what AI can do.”
When you buy through links in our articles, we may earn a small commission. This does not affect our editorial independence.
